241: Screening Efficiency Over Experience: Rethinking Cytology Expertise

Does more experience automatically make a cytotechnologist more accurate—or does where they look first matter more?

In DigiPath Digest #50, I review a digital cytology eye-tracking study that challenges the assumption that diagnostic accuracy improves steadily with years of practice.

The researchers tracked the visual behavior of 100 board-certified cytotechnologists with 1 to 40 years of experience. They found no statistically significant linear relationship between years of experience and diagnostic accuracy. Instead, low-power field efficiency—the ability to identify an important target quickly within a wider field—emerged as the key predictor of high accuracy discussed in the study.

The study also examined whether this visual skill can be developed. Twenty-eight students completed an intensive three-month cytotechnology training program. After training, they located diagnostic targets more quickly and spent less attention on normal, nondiagnostic cells. In other words, they learned both where to look and what to disregard.

What could this mean for digital pathology education?

As AI-assisted workflows take on more of the exhaustive searching, cytotechnologists and pathologists may increasingly work as expert verifiers. That requires rapid target assessment, strong knowledge of normal morphology, and awareness of risks such as confirmation bias and cognitive fatigue.

The study has an important limitation: it used static images rather than dynamic whole slide imaging. The findings raise useful questions about visual expertise, training, and competency assessment, but they shouldn’t be generalized beyond the study design without further research.

Episode Highlights

• 00:00 – Welcome to DigiPath Digest #50 and introduction to the paper
• 04:10 – Why the traditional definition of professional expertise is changing
• 07:02 – Moving from exhaustive searching to verification in AI-assisted workflows
• 09:04 – How eye tracking was used with 100 board-certified professionals
• 10:25 – Years of experience versus diagnostic accuracy
• 13:13 – Experience-based caution and attention to sample information
• 15:16 – Low-power field efficiency as a predictor of high accuracy
• 17:05 – Practical low-power field demonstration using a whole slide image
• 20:15 – Searching versus detecting and the mental map of normal morphology
• 24:04 – Comparing high- and low-performer visual scan paths
• 25:27 – Cognitive filtering: knowing what not to examine
• 27:35 – Can visual efficiency be taught in three months?
• 29:55 – How AI may shift the human role from searcher to verifier
• 30:38 – Study limitations: static images versus dynamic whole slide imaging
• 32:37 – Could gaze efficiency influence future competency assessment?
• 33:44 – Digital pathology learning resources and closing thoughts

Resources Mentioned

• Abstract and paper: Screening Efficiency Over Experience: Rapid Target Detection in Low-Power Field as a Modifiable Cognitive Biomarker for Diagnostic Accuracy in Digital Cytology

Listen to the full DigiPath Digest #50 recording to examine what the study found, what it didn’t prove, and how visual search skills could influence digital cytology training.

Transcript

00:00:00

 Morning. Good morning my digital pathology trailblazers. Let me know if you are tuning in again for the digipath digest. I almost didn't make it. I almost overslept. I slept I um woke up a lot later than I should for the for the live stream, but I am here and just uh looking for some of you in the meantime. I'm going to have a sip of coffee. I'm see I see you guys are joining. As soon as I have a critical mass of people, we are going to start. Let me show you what we're going to be talking

00:00:40

 about today. We are going to be discussing um this paper. It is called screening efficiency over experience. Efficiency over experience. How do we already feel about this? Um Can I play out talk? Can I I haven't done it for the whole summer. So now I'm again rusty with my um controls and where to click and everything. So bear with me if you are here at 6:04 uh in very very Pennsylvania. I was in Poland for quite some time. Then you are the real digital pathology trailblazer um and I appreciate you very much. So when you

00:01:37

 join uh let me know where you're tuning in from. I'm in PA in the US. Um when in Poland, we also visited Croatia. So let me know if you have visited any and places during the summer and how do you feel about going back to school? I have mixed feelings. I'm excited but I will also like not yet, please not yet. Um Okay, let's see which what can I do here? That's going to take me off a couple of clicks before you guys join. Um how do I make myself big? Uh Like this. Okay. I don't know.

00:02:38

 Then Okay, I see a few people joining. I know it's the first time. It's difficult for me as well, but if you're here, I appreciate you so much. So let's talk about screening efficiency over experience, rapid and target detection, um Let me use my We need to be a little bit interactive. All right. Screening efficiency over experience, rapid target detection in low power field as modifiable cognitive biomarker for diagnostic accuracy in digital pathology. Okay, first of all, our line is a lot too thick.

00:03:24

 Um but we can live with that. And Let's see if it's going to be the same when I switch to my presentation. Okay, I see more people joining. You did not disappoint me. Thank you so much, trailblazers. Um let me say hi in the chat. And uh let me know where you're tuning in from. I always want to know where you are based, located, and I very much want to know what time is it. Do you actually wake up for these at 6:00 a.m.? As some people do. I know they do. Like US-based people. Um, okay. So,

00:04:10

 how do you feel about it? Screening efficiency over experience. Like, I'm like, really? But, apparently, yes. Uh, so, this was published in Cancer Cytopathology this year in May, actually. Um, and the authors, Naoya Abe, Yuraki Nishimura, Kazuya Yashamita et al., they described a very interesting uh, phenomenon. And I love this like cognitive biomarker term as well. I've never heard that. Cognitive biomarker that you can modify. Um, this is very, very interesting. So, um, what happens? What's happening here?

00:04:57

 There the definition of experience, sorry, of expertise is changing. Because it used to be, okay, the more you practice, [clears throat] uh, the the longer you are active in your profession, the more experience you have, and the better you are, right? So, this is about um, cytology. We're talking about cytotechnologies. So, um, the old era they described as the manual endurance. Um, and and why endurance? [clears throat] Because it was very tied to physical endurance. Um, because you were scanning thousands of

00:05:43

 cells line by line over thousands of logged hours. And it would begin like like any pathology training, where you would sit with somebody who was doing it and probably on a multi-headed scope, and you would just like look with them. And then they would say, "Oh, malignant, benign." Or whatever diagnostic uh criteria they would use in um this particular case. Um and then you're supposed to like learn by osmosis. And I did that as well. Uh let me know if you did that because it uh was kind of a

00:06:24

 I don't know. Steep learning curve is kind of like when you learn a lot at the same time. To me, it was like, "Where is the curve? I'm looking. I'm like visually familiar with this, but I don't really know what to do with this and what makes this person sitting next to me decide one way or another." And then like at some point I started reading up and um additionally educating myself. Um so, the new era is uh a high-speed verification. So, what's happening is an an AI algorithm is

00:07:02

 flagging regions of interest and breaking the so-called 10,000 hour rule. What is this rule? This rule is that oh, you to master something, you need to do it for 10,000 hours. And then like people calculate, "Okay, how many day is that?" Uh if we had like a 10 day hour of doing something, that would be 1,000 days, right? Obviously. If you just do it for 1 hour a day, less days. And but there is a a different story now because you have AI flagging new things. So, the modern lab demands rapid, hyper-accurate

00:07:40

 human verification of machine-flagged anomalies, which is a difficult skill. Don't get deceived that oh, now like it's going to show you and you have to verify. There is a lot at play here um from like different areas that what I'm thinking about I'm thinking about confirmation bias, about fatigue, about um different things, right? Uh that are more um associated with a software giving you uh a potential answer and you having to decide. So, that's that's a challenge as well, but that's a new challenge. It's

00:08:21

 not like you're going to have to sit and do it for so many hours. Now you're going to have it shown and you have to decide. So, this part that I was mentioning, oh, I had to like read up, that's going to be crucial. Um but here's um what what what um they claim in this paper. Uh they're basically they're challenging the dogma of time. So, the dogma of this on-the-job training uh that assumes that competence accumulates linearly with years worked. Uh so, their university tested this

00:09:04

 hypothesis by by measuring something called tacit knowledge, uh which is the subconscious expertise professionals cannot easily verbalize. So, they worked with 100 board certified professionals who had 1 to 40 years of experience. And uh what they did in this paper, they were tracking their eyes. [clears throat] So, it was not like oh, tell me uh or I don't know know, whatever method they used, but they were basically doing eye tracking and um tracing where these people were looking on the screen. Where exactly? So, it's

00:09:50

 like um if uh >> [laughter] >> if you would compare it it's so funny that you know compare a human to an algorithm uh to a transformer or then it would be like the attention mechanism mechanism. Where does their attention uh stop for a second, right? So, hehe, I don't know. Let me know what you think about this eye tracking. I'm always like, "Oh my goodness." Not only can they track my mouse, now they can track my eyes. So, I'm very trackable. But, I guess we all are. We

00:10:25

 just uh are not aware of that. And let me know in the chat if you're still here. I see the numbers uh uh that people are joining, um but just say hi in the chat so that I know that everything is working, that you can hear me well, that you can see me well, and that you can see my presentation as well. Well, so, as we already stated at the beginning, let me make it a little bigger. Uh time does not guarantee accuracy. So, when you look at this chart, um at the bottom of the chart you have years of

00:11:01

 experience, so that would be 40 years. And uh here is the accuracy, 100% accuracy. That's not going to be ever achieved. And let's just look like where where the highest points are here. Um and this Yeah, so basically you have here high high just like let's just take the highest uh and this would be I don't know, 8 years, then 20 years. 30 years. So, it's not that like it goes everything clusters here. Nothing clusters here. There's nothing there. So, research found no significant linear

00:11:51

 color correlation between years of experience and diagnostic accuracy, which is crazy, I think, because it's always like, "Oh, they have so much experience. They must be so much better." But, I mean, no. No. Uh and it's not that like I want to give a specific example that oh, if you do uh like one specific thing for 30 years, it doesn't mean that you have done many different things, like [clears throat] one specific diagnosis, one specific type of um type of practice or whatever. It doesn't

00:12:25

 even have to do with that. Um they uh stated they they they discovered um that or uncovered that 3-year novices performed just as accurately as 35-year veteran. So, um basically, like this is their um Pearson's coefficient uh between accuracy and years of experience, and it was 0.189, which in statistics means it's flat. It looks kind of flat as well. So, well, what then does make the difference, right? Um and also um there was something they stated they I don't want to overuse the word

00:13:13

 discovered, but um noticed, noticed. Uh that there was something called experience-based caution, and uh that was not accuracy. So, uh what does that mean? And they couldn't know it from the eye-tracking data. It revealed that years of experience correlated with defensive medicine. Interesting term as well. Because veterans, those who were active in the profession of site technologists for over five years, they spent significantly more time reviewing sample information. So here is like the patient

00:13:54

 info um on the slide. And this is what they were looking at. How do you know? Well, from the eye tracking data, right? So you can see where So let's say this is the slide. And here's all the peripheral information. This is what they were looking at. >> [clears throat and snorts] >> One second. And um Well, it didn't help them. The staring longer at patient history did not improve their ability to identify malignant cells on the screen. So yeah, because it has nothing to do with

00:14:37

 malignant cells on the screen, but I can so relate to that because um I know that at the beginning it was like oh slide was the main thing to focus on and will I even recognize things in the slide and I would like go straight to the slide. Now when I review my studies, I review all the information available, which also is best practice, right? It's not that you should like leave this information out, but they just had this observation that okay, the more you practice, the more you looked at this information, but it

00:15:16

 didn't help you with your diagnosis. So what was the predictor of accuracy? That so predictor of high accuracy and it was something else. It was so-called low-power field efficiency. So they used multivariate analysis to strip away noise variables and the researchers found only one predictor of high diagnostic accuracy and this predictor was over 83%. This is crazy. And counterintuitively the most accurate diagnosticians were those who look at the target in the wide view the least. So I'm going to show you. Let's see if I

00:16:07

 can get a whole slide image to give you a real demo of this. But basically the low-power field efficiency like you look at the full slide and you immediately know where to go. So give me a second to open. Let's just do maybe we're going to go to path presenter. And let's see how how I how I will perform showing you this as an example. So just give me 1 second to log in. And Oh my goodness, don't give me Don't give me trouble. Okay, I'm just going to go to the slide library.

00:17:05

 And Okay. [clears throat] Let's see if I can share the screen. And >> Yes, I should be able to. Okay. You let me know in the chat that you see it. I'm just going to go to Okay, lung. Let's do a lung cancer. Um So, obviously, we're talking about cytology. I'm here showing anatomic pathology images, but basically what they mean by low power field efficiency, I already know I'm going in here and in these, right? This is probably my um What Do you see my mouse? You should be able to.

00:18:05

 Ah, no, it's not sharing the right tab cuz it's opening everything in the right tab. No worries. No worries. We can do it. We can do it. So, I already know I'm going in here. This is my mass. I see it from low low power field. Uh and these probably lymphoid aggregates. I don't care about this side. The These are vessels. This is lung tissue. So, I know we're talking about uh two different modalities for diagnostics, but basically this is low power efficiency. That on low power, you already know

00:18:46

 where to go in. Uh and I see myself freezing. I hope I can unfreeze. Um Let me know in the chat when I'm unfro- frozen, okay? Sometimes my internet looks high on the on paper. Okay, something is unfreezing. Am I visible to you? Okay, I am visible. I'm back. So, I don't know where we left off, but basically this would be my place to go into and see, okay, is it a tumor mass? And the answer is yes, by the architecture of the cells already. And then, okay, are these lymph nodes follicles?

00:19:29

 Uh yes, answer is yes. And I knew it from low power that I have to go there. And obviously, that's not the full diagnosis. It's just to show you how this works. Uh but that's what they mean by low power field efficiency. So, let's go back to our presentation. Yeah, that was the only thing. And the cool thing is, okay, so um is this good news or bad news? It's definitely good news because um if it's low power field efficiency, and you can like tell people where to look and how to do

00:20:15

 it, you don't have to wait 40 years for them to get enough experience. Um so, they distinguish here searching versus detecting. So, when you search, um you you would go like very linearly. So, you would go you would do this row, the first row of cells, second row of cells, third row of cells, and so on, right? But, they say experts uh they instantly take the entire slide relying on consolidated mental map of normal, which again brings me to, wow, this is what we're trying to teach the foundation models or the anomaly

00:21:02

 detection models. This is what we are doing already, the cytotechnologists in this paper. They basically look at the whole thing and they recognize, oh, there is a not normal dot. And what they're doing, they like filter all this out, all the stuff that's around it out, and they just go straight to that cell that is so looking so so funky. And believe me, people, pathologists talk like that. Something looks funky. They say it's funky. So, they Whoa, I just splashed coffee on myself.

00:21:45

 So, they filter the noise. This is called parafoveal pop-up, meaning that abnormal abnormalities disrupt the mental map and instantly pop out in peripheral vision, bypassing active searching. So, it's like different part of physiology that you look for searching for things. And then step three, obviously, is high power verification, and the expert acquires the target instantly and switches to the zoomed-in view. So, I want to show you something. Oh, no, I'm frozen again. I hope you're not leaving me because I'm

00:22:28

 frozen. It's so funny that this is what's happening in the diagnostic process. So, Let's see. You're seeing it. Okay. You should be seeing >> [sighs] >> That's a I'm waiting for me to unfreeze. It's a little difficult to work like that, but Okay, I'm I'm frozen. So, this high power verification, right? We have low power Where is my Here. We We have low power and I immediately go in here. Right? And look at the And then I cannot scroll fast enough. Then I go back, search

00:23:20

 something and let's see here. And it is constantly in out in out and believe me, if you're not a pathologist and you're not doing this for a living, it can be very very annoying for the observer. Just switching windows here. And I know that cuz sometimes my husband we have a home office together and he looks at me looking at slides and I want to sometimes show him something or like ask him about things in bone marrow specifically cuz he's a clinical uh pathologist and I do that. I zoom in out

00:24:04

 zoom in out and he's like, "Stop. I need to look at it." And so, yeah. But that's how diagnostic process works. Mm, so let's talk about the timelines. And the eye tracking software mapped visual scan paths second by second on the exact same slide. And if [clears throat] you look at this, the high performer's path was a lot straighter and straightforward than the low performer path. So, high performers needed 40 seconds and whereas low performers 104. The style was linear and

00:24:47

 rhythmic and the style of the low performer was chaotic and fragmented. And And action was direct shift from low to high. So, it like zoom in and we're out. Zoom in, zoom out. Zoom in, zoom out. Zoom in, zoom out. So, a direct shift from low power field to high power field verification. Uh whereas the low performers, they were searching and recursive checking of the empty background. Um that's what happened, but there are a couple of caveats that we're going to mention in a second. Mhm,

00:25:27

 here the the important thing to increase your efficiency is to know what not to look at. So, the visual expertise is not merely knowing where to look. It is the strategic disregard regard strategic regard of visual noise. And what is visual noise? Everything that is not your kind of like target that you're looking for. Uh it cells blood inflammatory inflammatory cells there may be your noise. And and this is called it has an official name called uh and this is cognitive filtering this protects the

00:26:12

 cognitive filtering bandwidth. Um and the sad thing is that unstructured on-the-job training without objective feedback accidentally reinforces the exhaustive line-by-line novice scanning habit. Because if you don't um tell why this one is not abnormal. Uh we like intuitively know here the other dots are gray and this one is black. That's why it's abnormal and you immediately see it. But the um cytopathological features are a lot more subtle. So, then the training would be okay, I went in here because these are

00:26:58

 the features. So, that was my like okay, when I read up on it and I studied normal histology and studied the abnormal. Um you do need to know normal to disregard it. So, that's also a requirement. So, the million-dollar question or I don't know how many-dollar question, but education and healthcare education, they're they're they are areas hopefully of significant investment is, can cognitive efficiency be taught? If 40 years of unstructured experience won't guarantee pop-out detection, can

00:27:35

 it be artificially manufactured? And the answer is yes. And it took these researchers 3 months. Can you imagine? Well, they enrolled 28 students in an intensive 3-month cytotechnology training program. And they evaluated via matched set 30 diagnostic images before and after the program. So, um before you ever trained, uh you have a set of images and after you trained. And the goal was to measure objective visual search skill acquisition over a compressed timeline. That is amazing, 3 months. Um and they did validate the rapid skill

00:28:22

 acquisition. Um so, the time to first fixation on targets decreased substantially. So, basically they faster spotted everything. Um and the time to first fixation on normal non-diagnostic cells increased. So, strategic disregard. So, they increased spotting these cells faster and they increased the disregard of background. So, they the students gained both skills, what is the abnormal and what is the normal measured by okay, faster spotting abnormal and uh disregarding the normal. So, students actively learned to ignore the

00:29:09

 background acquiring expert-level selective attention in just 3 months. This is huge. This is huge, my friends. This is huge, trailblazers. Like, just by teaching differently because you know what's relevant, you can decrease this to 3 months. This is amazing. So, um they became verifiers and not searchers. In digital pathology, this is So, there's a whole um area of science. It's the computer user interface science. There's like It's a science. Uh like how the user interface is in

00:29:55

 computer and how people interact with computer. So, digital pathology is kind of part belongs to that area. And now, uh AI now handles the exhaustive searching. And the future diagnose ability to rapidly and accurately verify AI suggestions. Um which is a different skill, right? And non-employing professionals uh stuck in a serial search mindset will suffer massive workflow drag. So, they're not saying that the people who like did it for 40 years, they were still doing the serial search field by

00:30:38

 field by field by field. Uh but what they're saying is this is how you start, and if you can um compress it, like eliminate it and look in a different way, that's fantastic. So, AI is going to do the exhaustive search for you, and then human diagnostician will will do the rapid verification, and we have an optimized lab output because of that. But there are boundaries, there are limitations. Um, the study utilized static images rather than dynamic whole slide imaging. Uh, so what I showed you on PathPresenter

00:31:16

 was dynamic. Uh, here they used static images. Um, what they say about that is actually uh it's it is a limitation, but it isn't because when you uh look at uh a certain field of view in a whole slide image, you have it as a static image while you're you looking at it. Um, so um that's what they say that even when navigating dynamically, uh zooming in and out, uh human feature extraction still relies on static field of view interpretation. Um, so the cognitive mechanics of pop-out detection operate independently

00:31:56

 of specific slide preparation. So, um a conventional slide with smears or liquid base, um also now um at least for um hematopathology, uh which is also like cell-based instead of tissue-based, you get a bunch of fields of view. So, it's uh you don't have to like look through a slide, but you have rows of cells shown to you. So, basically uh that's another skill to acquire, which hopefully um I mean not hopefully, gladly can be acquired in short period of time. So, also um that makes me happy because I'm like,

00:32:37

 "Okay, that means you can learn working in this new way fast. It's not that you now have to have that a decade of experience in digital pathology to start working in digital pathology." No, this is physiology, new way of interacting with the machine, and you're good. In this study, they showed 3 months and you have expert level performance. So, the question that they're asking in this publication is, "What about the future of certification?" Well, now cumulative time-based training

00:33:08

 is becoming obsolete for digital diagnostics, and the low power field efficiency is a quantifiable, modifiable cognitive biomarker. And here I want to emphasize this modifiable. That means you can teach it. So, how long until professional boards, medical boards, require objective gaze efficiency scores to prove readiness for the digital lab? And that both like makes me happy and scares me that they're going to be actually like making sure I'm looking at the right places. Makes sense, but still it's like

00:33:44

 feels kind of invasive into the way I work. So, um yeah. We are back after after uh a break, my trailblazers. And if you don't have the book yet, get the book. Um can I The book is on the digital pathology place website. You can get it there immediately. I'm going to share the website on the screen. When it loads. When it loads. Um but this book is for those of you who are starting your digital pathology journey, digital pathology one-on-one, all you need to know to start and continue your digital pathology journey.

00:34:39

 What do I mean by continue? If you start in a specific area, there is a lot about all the other areas, and me myself, I started in the area of image analysis, and here there is a lot more than image analysis, and this book is available for you for free from the website. Or from my store. Oh my goodness. I cannot even find where my own book is. That's okay. When you go to digitalpathologyplace.com, you're going to find it, and next time I'm going to prepare a QR code, and actually what I'm going to do, I'm just

00:35:25

 going to drop this in the chat for you. And other things that we're working on, if you want to learn what is normal, um because like they showed in the study, uh the key to being fast with the abnormal, you also need to know what normal is. So, you need to like disregard the things that is are background for the particular task at hand. Uh then we have uh um histology series. Uh I'm teaching you normal histology on real whole slide images uh in the live stream format as well, and we're going

00:36:06

 to be having one on Friday. Uh when you get the book, you're going to know about this live stream as well, and all the other things and that I'm doing online and offline, the conferences I'm going to be going. I'm so excited to be back with you, my trailblazers. If you have any digital pathology questions, let me know in the chat even if you're watching the recording. Another couple of things that are happening. So, still working on the second edition, just images, just images. And I just need to start life

00:36:43

 comes in between, but it's going to happen. And another thing I am thinking of is well, I already kind of started working on it is revising the digital pathology club, which is a membership where we dive deep into publications when there are courses to take you from zero to digital pathology Don't say here. I want to say competent professional digital path from zero to digital pathology competent professional in like definitely three months, you could do that in the membership and then maintain your

00:37:23

 knowledge with reviewing literature together and seeing what's happening uh in the digital pathology space. So, that's going to be out there soon. I'm going to let you know. And other than that, thank you so much for joining. I as always, I appreciate every single digital pathology trailblazer.